Michael Shehata
about
Michael is a Bentley University Trustee Scholar, selected as one of 10 from 10,000 applicants. He ranks in the top 1% on TryHackMe and received a $6,000 Oberlin scholarship for cybersecurity research with Professor Hanan Hibshi at Carnegie Mellon. He was also selected from 10,000+ candidates to serve as a Python instructor and researcher at Stanford University (top collaborator among 80 researchers).
He reviews for top‑tier AI venues such as COLM 2026 and the ACL REALM 2025 workshop and has four peer‑reviewed machine learning papers. He also placed 44th of 50,000 in the Climate Science Olympiad using robotics and ML for plastic‑pollution mitigation, and engineered an underwater ML system that won 1st place globally at MATE ROV 2023.
now
- Bentley University — NSF‑funded robotics haptics research with Prof. Mounia Ziat.
- getinlightened — data engineering on pipelines and analytics infrastructure.
selected research & projects
- Few‑shot LLMs with dynamic prompts — BERT cosine‑similarity based selection of in‑context examples for each query; 100+ experiments on math‑reasoning benchmarks, +8% over Chain‑of‑Thought with text-embedding-3-large. COLING 2025.
- Conversational acuity in language models — first‑author work using pragmatic measures to score and re‑rank candidate responses from LLMs, yielding a 34.9% improvement in conversational quality over baselines. ICTC 2024.
- Autonomous robotic manipulation — imitation‑learning policies trained on a 50k‑frame RLBench dataset for robotic arm control, reducing human intervention and improving motion accuracy. ICMRV 2024.
- Driver‑monitoring system — CNN‑based pipeline for yawn and eye‑closure detection on TI‑TDA4VM edge hardware with 89% accuracy under varied lighting. Int’l Mobile, Intelligent, and Ubiquitous Computing Conf. (IMUCC 2025).
- Quantum approaches to text — quantum‑inspired representations for language with IYRC STEM (Columbia, Stanford, Harvard).
- MATE ROV control & vision — graphical control interface + YOLOv5‑based object detection (96% accuracy) for a $3,500 underwater robot; 1st in computer vision & programming, plus multiple design and report awards.
More work is on github.com/michaelnaeim.
publications
- Enhancing LLM Conversational Acuity Using Pragmatic Measures. 15th International Conference on Information and Communication Technology Convergence (ICTC 2024), IEEE.
- Reinforcement Learning for Programming Feedback: Aligning Small LMs Without Human Preferences. SIGCSE 2026, ACM.
- Optimizing Driver Monitoring System Performance on TI‑TDA4VM Edge Device. International Mobile, Intelligent, and Ubiquitous Computing Conference (IMUCC 2025), IEEE.
- Advanced Machine Learning in Robotics. 10th International Conference on Robotics and Machine Vision (ICMRV 2024).
reviewing
Peer reviewer for COLM (Conference on Language Modeling) 2026.
Peer reviewer for the ACL REALM Workshop 2025, focusing on language modeling, retrieval, and evaluation.
work
ML research: 100+ experiments on few‑shot learning and multilingual retrievers; 8% gains over Chain‑of‑Thought on math tasks with text-embedding-3-large.
Cloud / DevOps: 25+ analytics dashboards; AWS S3 pipelines over 12k events.
Cybersecurity research with Prof. Hanan Hibshi on CNN‑based SQL‑injection detection using hybrid approaches.
Robotics research with Prof. Ke Wang (Imperial College London) on ML for robotic manipulation; paper at ICMRV.
teaching
Python instructor and research collaborator for the Code in Place AI Evaluation Project. Led weekly CS106A sections (14 learners) on Zoom and built automated grading and ML‑driven feedback for 5+ labs.
Founded an affordable robotics program; taught 240+ students and led 15 teams in the International Robot Challenge.
skills
Python, C, C++, Java, JavaScript (React, Node, Vue), SQL, Assembly, PHP, Linux, OpenCL, PyTorch, TensorFlow, Keras, scikit‑learn, Pandas, NumPy, Matplotlib, Django, Docker, AWS, Git, PCB design.
selected awards
- Bronze Medalist, Climate Science Olympiad — 44th of 50,000; invited to COP28 for robotics & ML work on plastic pollution.
- MATE ROV 2023 — 1st in computer vision & programming; best float design, report, and presentation.
- Kraza Full‑Stack Web Dev — 1st of 8,000 (2020), 2nd (2021) in a national competition building church websites.
- ICPC ECPC Teens — top 80 of 2,650 teams; Meta Hacker Cup — top 20% of 27,000 competitors.
- Finalist, 18 Under 18 (Crimson Education); peer reviewer for COLM 2026 and ACL REALM 2025.
- Top 1% of 1.9M users on Keybr (120 wpm) and top 18% of 3M on TryHackMe.